Do Musicians Dream of Electric Violins?

UoM administered thesis: Phd

  • Authors:
  • Thomas Lloyd

Abstract

Yes. While the results presented in this thesis declare that listeners prefer the recording of an acoustic instrument, there is still a positive response about virtual violins, especially when compared to an unfiltered electric instrument. In this context a virtual violin is the result of convolving the raw signal from an electric instrument with the impulse from a real violin. A key part of this process is the characterisation of this impulse response, of which there are several methods. This thesis explores the use of virtual violins in acoustics research and music performance. The opening chapters provide an overview of the literature about violin acoustics and previous uses of virtual violins. A significant portion of the thesis details the development of a system that is used to produce digital characterisations of violins. The method used involved the measurement of sound radiation from the violin body after it had been excited by an impulse. This impulse is provided by an instrumented hammer, which strikes the violin on the bridge. One of the pitfalls of this method is the imperfect frequency response of the strike, which is corrected using a deconvolution algorithm. Deconvolution is an important part of the process and is discussed at length in the thesis. The described characterisation system utilised a bespoke frame that could hold the violin by the neck and rotate it to a specific angle on a single plane. This enabled the characterisation of a violin at incremental angles. Not only can these characterisations be used for the development of virtual violins, but they can also be used to analyse the spectral properties of the instrument. These characterisations at different angles allowed an examination of the violin's directivity, which explored how the sound energy is distributed from the instrument. Analysis determined that in the mid-range (C4-B5) the sound distribution is fairly isotropic. Outside of these ranges the sound distribution is significantly anisotropic. Finally, the thesis details three psychoacoustic experiments where participants listened to different audio samples and rated them according to their personal preference. The first of these had five virtual violins, of various manufacturers and ages, along with an unfiltered electric instrument. Listeners of mixed musical ability were asked to listen to recorded samples of these instruments and to provide preference scores for each. The experiment found that listeners preferred the sound of virtual violins to the unfiltered electric. The next experiment presented musically trained participants with samples of a virtual violin with varying impulse response lengths. This revealed that there was a sigmoid shaped relationship between the number of coefficients of the violin's impulse response and the listener's preference score. The final experiment was designed to examine the preference scores towards a recording of a real violin and its emulated equivalent. Participants of two groups, one with musical training and one without, took part in the study. The results showed that the real violin was preferred to the emulated violin and that musical training did not have any effect on the preference scores. The statistical analysis also determined that there was no interaction between violin treatment and musical training. An additional question was also posed to the participants post-experiment: Which of the presented samples is the real instrument? The results showed that those with musical training were significantly better at identifying the real instrument compared to those without such training. Speculation is drawn as to why this might be, though it is speculated that the spatial effects, such as those described by Weinreich, had no effect.

Details

Original languageEnglish
Awarding Institution
Supervisors/Advisors
Award date31 Dec 2018